Data-Center Opposition Is Pushing AI Companies Toward Local Deals

More than a dozen Ohio towns voted to halt or restrict new data-center development in the midterms. Lon Harris also cited similar bans in Kansas, Florida and California, and said Texas, Georgia and Ohio account for more than 40% of planned US data-center capacity. Texas governor-elect Gina Hinojosa campaigned on pausing construction until the legislature passes new rules.

The results put a practical question ahead of an abstract one: what conditions will communities require if data centers are to be built nearby? Jason Calacanis argued that local opposition did not make construction impossible. Instead, operators would have to address residents’ concerns about electricity, water, pollution, noise and the benefits towns receive.

Calacanis had predicted in August that candidates might campaign against data centers and then, once elected, negotiate with operators. In the clip Harris played, he imagined officials saying they had “made peace” with data-center companies, which would pay for electricity and contribute to local amenities. Calacanis called the prediction “nailed it” after watching it back.

A post by investor Gavin Baker offered a specific version of that possible bargain. Baker wrote that Hinojosa had said currently permitted Texas data centers would be allowed to finish construction, while future projects could face requirements involving taxes, consumer power prices, water availability, union labor and local input. Harris read out those conditions; Calacanis said they sounded reasonable.

For Calacanis, the route to local acceptance was to make commitments concrete: pay taxes, create jobs, protect water and electricity access, avoid pollution and contribute to the community. He suggested that companies could compete to offer towns more, pointing to Meta’s willingness to improve its offers to communities. Being a good neighbor, he said, also meant paying attention to noise, smells and energy use—not treating a data center as acceptable simply because it promised economic growth.

Harris thought the debate had become too closely tied to party allegiance, with Republicans expected to support data centers and Democrats to oppose them. Calacanis agreed that the issue was not inherently partisan. Both emphasized local costs and protections, though Calacanis made the case more forcefully: in his view, operators should explain repeatedly how a project would affect electricity bills, water use, taxes and jobs.

That emphasis on communication informed Calacanis’s response to OpenAI’s new commitments. Harris said the company had pledged to pay the full cost of powering facilities it owns and operates, seek community input, publish construction plans and avoid nondisclosure agreements with public officials. A separate policy blueprint encouraged lawmakers to enact similar commitments. Harris also noted that Meta, Microsoft, Google and Anthropic had outlined comparable commitments, and that OpenAI had told investors it was targeting $600 billion in total compute spending by 2030.

Calacanis welcomed the commitments as proactive outreach, but did not describe them as revolutionary. He argued that companies need to listen to residents, reflect their concerns back to them and explain their plans repeatedly. The message he thought operators should keep delivering was specific: a facility will not affect local electricity bills, will use recycled water or less water than nearby golf courses, and will pay its taxes. Repetition, he said, is part of leadership because it increases the chance that people understand and remember what is being promised.

Those commitments sit alongside a large planned buildout. Harris reported that senior OpenAI leaders were discussing whether to reset expectations for next year’s capacity by Wednesday morning. He also cited a Bloomberg report that Microsoft was souring on neoclouds such as Nscale, Iris and Nebius, and had canceled a roughly $3 billion Nscale deal tied to a UK data center. The local fight over the conditions for building and the industry’s plans for computing capacity were unfolding at the same time.

Democrats secured majorities in both the House and Senate. Calacanis described the midterms as a referendum on Donald Trump, while giving him an A to A+ for supporting Big Tech and the AI buildout. In his view, Trump had made data centers a focus, stood his ground when they became unpopular and urged communities to negotiate rather than reject projects outright.

Harris asked whether Trump had made data centers more partisan by putting himself at the center of the issue. Calacanis agreed that Trump had turned the topic into a wedge issue, forcing Democrats toward the opposing position. He expected elected Democrats to move toward the center when governing, because they would also need jobs and economic growth. That was his forecast, not a conclusion established by the election results.

The hosts described several different pressures on voters. Harris cited the Iran war, opposition to AI and data centers, and concerns about corruption. Calacanis put affordability first and connected it to the war, while agreeing that concerns about democracy and self-dealing mattered. He said he did not want to prejudge allegations of grifting and would wait for investigations to establish what happened. The discussion offered competing emphases rather than a single account of what drove the result.

Harris described Democratic leaders as having limited procedural leverage and choosing to wait for voters to judge the administration’s policies, including tariffs and the Iran war. Calacanis called the approach “do nothing.” Harris’s account was that leaders such as Hakeem Jeffries and Chuck Schumer judged that they had little practical power to stop the administration and that voters would be able to assess its record at the midterms.

Calacanis’s own account of the administration was mixed. In clips Harris played, he criticized Trump’s memecoins and the chaotic rollout of tariffs, while also defending the case for addressing trade imbalances. He said the Trump family had not been sent to Washington to launch a memecoin, even if family members were free to pursue private business. On tariffs, he said a more deliberate approach would have grouped trading partners by importance, assigned officials to negotiations and prepared the public for likely effects, rather than escalating through public declarations.

Calacanis argued that Trump could be right about a problem and still execute the response badly. He said tariffs were important in addressing trade imbalances, but that negotiations could have proceeded without public spectacle. He also said advisers should have been willing to tell Trump when his approach was not working. In his account, a circle of people afraid to contradict a powerful leader made it harder to change course.

To explain how he assessed the tariff effects, Calacanis described speaking with founders, including the CEO of Cut Clothing, about what the policy was doing to their businesses. He said he preferred to hear from people directly affected rather than rely on politicians or commentators. Harris noted that the founder was a Trump supporter, making his account a counterpoint to the assumption that criticism came only from political opponents.

The market moves offered another signal, but not a settled one. The source description reports that Big Tech added nearly $1 trillion the day after the election. Calacanis described a subsequent 5–6% correction in tech stocks. He also said investors might be anticipating a check on Trump over tariffs and the Iran war: if tariffs calmed down and the war ended, he argued, that would be bullish for stocks.

The correction did not change Calacanis’s view of the technology trade. He called the selloff overblown and described the moment as a generational buying opportunity, naming Nvidia, Amazon, Uber, Tesla and SpaceX as stocks investors might consider. Harris said he had bought some shares during the dip. He described waiting for companies he considered fundamentally strong to fall in price as the strategy that had worked for him.

Calacanis’s forecast that political resistance would lead to negotiation rather than a halt to construction was consistent with his earlier prediction about officials making peace with data-center companies. But the votes, the market rally and the later correction do not establish one settled verdict on the buildout. They show, in the account presented here, a politically exposed industry still attracting investment while communities press for conditions on local projects.

The dispute over data centers is also a dispute over who benefits from technological growth and who absorbs its costs. In a clip from May, Calacanis described tensions inside the Republican coalition over automation, jobs and the concentration of wealth. He named truck drivers, cab drivers, delivery drivers, factory workers and Amazon employees as people whose jobs could be eliminated or substantially reduced, and said they would need to be redeployed.

Calacanis connected those fears to the possibility that technology companies could accumulate enormous value while workers were left behind. He framed the conflict as one between “America First,” workers’ interests and the power of what critics call tech oligarchs. Harris argued that economic populism could speak to those concerns more directly than culture-war disputes: candidates could keep returning to billionaires, wages and the minimum wage rather than letting opponents define them through issues such as pronouns.

Calacanis also urged Republicans not to characterize the entire left as socialist extremists. He said many of the candidates who won were moderates and should work with the center. Harris agreed that candidates could focus on economic populism rather than culture-war issues. Their discussion pointed to a political tension in the AI debate: assurances about local power bills and water use address the effects of individual projects, but not the broader question of how automation’s gains and losses are distributed.

The same distribution question surfaced in their discussion of wealthy people borrowing against assets. Harris said voters may focus on assets and equity because they believe wealthy holders can borrow against stock rather than receive salary income. Calacanis objected to the prospect of very high income-tax rates and said he would consider retiring if taxes reached 60%. Harris suggested that the political argument might focus more on holdings than income, since income taxes fall on people working day to day while some wealth is held in appreciating assets.

Calacanis proposed taxing margin loans above a threshold, suggesting $10 million as an example. He said he had used a margin loan to buy a ranch and acknowledged the appeal of borrowing against equity. His concern was that someone could borrow to fund living expenses while an asset continued to appreciate, deferring taxes for years. The threshold was offered as an idea, not as a developed policy plan. The discussion connected public anger about unequal taxation to the question of how economic gains are realized and shared.

AI also has a direct cost inside companies: the tokens used to run models. Harris reported that Apple had lowered the daily AI spending cap for engineering teams from $300 to $200 in October, with further cuts in the works and some allowances falling to $100 a day. Outside engineering, the company was reportedly weighing a daily limit as low as $25.

Calacanis treated the numbers as a sign that companies may have to move beyond flat-rate subscriptions as use increases. He said his team paid $40 a month per person for Grok, but estimated that heavier users might consume $1,000 or $2,000 worth of tokens a year. Flat pricing, he argued, is designed to get people using a product; as usage and value rise, corporate plans could cost $100, $200 or even $300 a month per person.

He expected AI subscriptions to become a $3,000–$5,000 annual expense, comparable to or greater than many software licenses. If a service raised its price from $40 to $100 a month, he said, he would first look at who was using it and consider removing inactive seats. He also suggested that companies might share logins for light users, while acknowledging that he did not know whether this would violate a service’s terms.

Calacanis saw Apple as a special case, suggesting the company might arrive later than competitors but deliver a strong product, potentially using its hardware to run local models. Harris said a model-agnostic approach—letting customers use cloud models or run models on their own devices—made sense for Apple. Their discussion left open how businesses would divide the expense: absorb it as a subscription, pass it through to employers, or reduce it by running models locally.

The costs of AI are not limited to electricity and model use. Harris reported that Anthropic had struck a content-licensing agreement with the Associated Press covering election coverage. The financial terms were undisclosed; Claude users asking for polling updates would receive current information attributed to the AP. Harris noted that OpenAI had already made deals with publishers including Axel Springer, News Corp and The Atlantic.

Calacanis argued that using a publisher’s reporting in a competing, monetized service was not fair use. He distinguished the use of current reporting in a product from the separate question of whether material can lawfully be used to train a model. That training question, he said, remained unresolved. These were Calacanis’s legal assertions, not a determination established in the discussion.

He distinguished several issues: use of pirated source material; access that may violate a site’s terms of service; whether lawfully obtained material can be used for training; and whether a model’s output reproduces protected work. Harris clarified that Anthropic’s book settlement concerned bootlegged books, not a definitive ruling on training. Calacanis agreed that the training question was still open.

His proposed answer was broad licensing rather than prolonged litigation. He suggested that a highly profitable AI company could devote a portion of its revenue to media and training-data providers. Even a pool of several billion dollars a year, he argued, could matter to newspapers, magazines, books and online writers. He floated the possibility that a platform such as Substack could distribute a share to participating writers, perhaps with an opt-out system.

Harris asked whether Anthropic had made the AP deal partly to reduce legal and reputational risks before a possible IPO. The question was raised but not answered. The agreement nevertheless illustrated a practical distinction in the discussion: a company might maintain that model training is lawful while separately licensing current reporting used in a product that delivers news to users.

Across the discussion, the hosts returned to a difference between opposing AI in principle and negotiating over its effects. Calacanis argued that data centers could proceed if companies addressed power, water, pollution, taxes and community benefits. Harris emphasized that the debate had become politically polarized and that the local questions were more concrete. OpenAI’s commitments, and the conditions Baker described for future Texas projects, offered examples of how companies and officials might make those questions explicit.

Calacanis’s confidence that the industry could negotiate its way through local resistance was a forecast, not a guarantee that every community would accept a project. The towns’ votes and the state-level debate made clear that opposition had become consequential. Meanwhile, investment in AI remained substantial, even as the hosts discussed costs to residents, workers, companies and content publishers. The unresolved issue is not simply whether AI infrastructure will expand, but how clearly its costs and benefits will be assigned.